BREAKING
Supertonic 3: 99M On-Device TTS
0
M
parameters
0
languages
0
platform SDKs
Throughput by hardware (chars/sec)
M4 Pro CPU
1263
WebGPU
2509
RTX 4090
12000
On-device vs cloud TTS
Supertonic 3
local
●
Runs offline, no GPU
●
Reads dates and units reliably
●
Open weights, MIT sample code
Cloud TTS
metered
●
Per-character costs
●
Network latency
●
Privacy trade-offs
Quality-speed trade-off
2
Coarse, slurred
↓
5
Usable quality
↓
–
Throughput drops
Speech that stays on device
AI NEWS BLITZ
Supertone has released Supertonic 3, a tiny speech model that runs offline on a Raspberry Pi.